Indoor Human Activity Perception Based on Infrared Thermopile Array Sensor

Human activity perception systems play a significant role in “SENSEable City”, smart homes and medical care, etc. Thermopile infrared array sensor (TPAS) is a new type of sensor to detect human targets without privacy invasion. This paper focuses on the perception of static and dynamic human activities. A hardware platform is built based on TPAS MLX90640 with the output of $24 \times 32$ pixels, to collect the thermal data of human targets from both side view and top view. After temporal Gaussian filtering, background removal, motion information extraction and data enhancement, different improved convolution neural network structures and technologies are introduced for activity perception. The best model among these networks for static and dynamic activity perception is selected and trained, achieving perception accuracy of 97.5% for static activity and 94.6% for dynamic activity respectively. The experimental results have shown that the system we proposed could correctly perceive continuous human activity and meet the real-time requirement as well.